A Study Of Urban Road Intersection Extraction From Remote Sensing Image

Road is not only an important military facility, but also an important civilian facility. Remote sensing image can cover a wide range of area and can be gotten easily and timely. If the road can be extracted correctly from the remote sensing image, it is very useful for designing battle plan, update of foundational geographic data. Also it will accelerate the developing of vehicle automatic navigation technology. However, due to the character of urban area, the research on how to extract the road from the urban area is few. But people pay more attention to the rural area and suburban area. This thesis will study the factors that obstruct the road extraction from the urban area and offer a new method to extract road intersection from the urban road network.In this thesis, after the first analysis and experiments, choose the median filter to filter the salt and pepper noise which is in the remote sensing image. At the same time, use the mathematics morphology to clean out the vehicles, pedestrians and direct lines which are on road. Its purpose is to make the gray of road become smooth and the road extraction more easily. The key that obstructs the road extraction form urban area is the intersections of the urban road network. So, if the intersection could be extracted, it likes finding the urban road network. Therefore this thesis's main content is to extract the intersection. Based on the research of visual recognition mechanism and memory mechanism, this thesis pays more attention to the process of visual recognition. In this thesis, use it as the guide to design the intersection extraction algorithm.After the analysis of the characteristics of road in remote sensing image, find out the image features that can be used as the stimulation of the algorithm. Extract the number of gradient peak for the strong directional character of the road edge. The peak number stands the number of road in the current window. On this basis, combining with the road gray and texture features, pre-extract the road intersection from remote sensing image. The aim is to quickly rule out the non-intersection windows and give the weight to the suspected windows in accordance with prior knowledge.In order to use the shape feature, this thesis obtains the requirements by an ideal test and designs a labeling procedure on the basis of them. As the labeling method is similar, design a thinning procedure to correct it. To this end, this thesis firstly uses the shape feature to extract the road intersection. In addition to some shape parameters, this thesis also uses intersection branch feature to extract the intersection. By finding the number of the branch and the location, direction of every branch, make the road intersection extraction algorithm more accurately. At the same time, record them as the members of the intersection class. Finally, based on the process of visual recognition, integrate the above methods to design the road intersection extraction algorithm. The experiment also proves that the algorithm can extract road intersection form urban area accurately.